
{"id":9598,"date":"2026-06-25T05:12:22","date_gmt":"2026-06-24T21:12:22","guid":{"rendered":"https:\/\/infernews.com\/blog\/biomatrix\/"},"modified":"2026-06-25T05:12:40","modified_gmt":"2026-06-24T21:12:40","slug":"biomatrix","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/biomatrix\/","title":{"rendered":"BioMatrix \u628a\u751f\u7269\u5e8f\u5217\u8207 3D \u7d50\u69cb\u653e\u9032\u540c\u4e00\u6a21\u578b"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-62e28701749b.jpg\" alt=\"BioMatrix\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">BioMatrix \u662f\u4e00\u500b\u591a\u6a21\u614b foundation model\uff0c\u5efa\u7acb\u5728\u55ae\u4e00 decoder-only \u67b6\u69cb\u4e4b\u4e0a\u3002\u5b83\u8981\u89e3\u6c7a\u7684\u554f\u984c\uff0c\u662f\u628a molecules\u3001proteins\u30011D sequences\u30013D structures \u8207\u81ea\u7136\u8a9e\u8a00\u653e\u9032\u540c\u4e00\u5957\u751f\u6210\u6d41\u7a0b\uff0c\u4ee4\u6a21\u578b\u4e0d\u53ea\u53ef\u8b80\u53d6\u4e0d\u540c\u8cc7\u6599\uff0c\u4e5f\u53ef\u7528\u540c\u4e00\u500b next-token prediction \u76ee\u6a19\u8655\u7406\u8207\u8f38\u51fa\u5b83\u5011\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u6709 biological foundation models \u901a\u5e38\u5206\u6210\u5169\u985e\uff1a\u4e00\u985e\u53ef\u5728\u5171\u4eab\u76ee\u6a19\u4e0b\u878d\u5408\u591a\u6a21\u614b\uff0c\u4f46\u591a\u6578\u53ea\u96c6\u4e2d\u55ae\u4e00 entity type\uff1b\u53e6\u4e00\u985e\u96d6\u7136\u8986\u84cb molecules \u8207 proteins\uff0c\u537b\u5e38\u5e38\u6b20\u7f3a\u986f\u5f0f structural modeling\uff0c\u6216\u8005\u4f9d\u8cf4 adapter-based designs\u3001external encoders\u3001projection adapters \u8207 modality-specific output heads\u3002BioMatrix \u7684\u53d6\u5411\u5f88\u9bae\u660e\uff1a\u76f4\u63a5\u628a SMILES\u3001SELFIES\u3001\u5206\u5b50 3D\u3001\u86cb\u767d\u8cea\u5e8f\u5217\u3001\u86cb\u767d\u8cea 3D \u540c\u81ea\u7136\u8a9e\u8a00\u6620\u5c04\u5230 shared discrete token space\uff0c\u5c07\u300c\u53ef\u8b80\u300d\u8207\u300c\u53ef\u751f\u6210\u300d\u7d71\u4e00\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6280\u8853\u4e0a\uff0c\u9019\u500b\u9805\u76ee\u6700\u503c\u5f97\u7559\u610f\u7684\u662f unified tokenization scheme\u3002\u5206\u5b50 3D \u7528\u6539\u826f\u7248 MolStructTok\uff0c\u86cb\u767d\u8cea 3D \u7528 GCP-VQVAE\uff0c\u4e26\u4ee5 description-based embedding initialization \u628a\u65b0\u589e token \u5148\u5c0d\u9f4a\u5230 pretrained Qwen3 embedding space\uff0c\u518d\u505a continual pretraining\uff1b\u9019\u7a2e\u505a\u6cd5\u6bd4\u8d77\u5f8c\u52a0\u6a21\u614b\u63a5\u982d\u66f4\u5b8c\u6574\uff0c\u4f46\u8a13\u7df4\u6210\u672c\u4ea6\u660e\u986f\u66f4\u9ad8\uff0c\u5b98\u65b9\u8cc7\u6599\u63d0\u5230\u66fe\u7528 64 \u5f35 NVIDIA H100 GPUs \u914d\u5408 LLaMA-Factory \u8a13\u7df4\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f9e GitHub \u8207 Hugging Face \u73fe\u6709\u8cc7\u8a0a\u770b\uff0c\u9019\u500b\u9805\u76ee\u8f03\u9069\u5408\u7576\u4f5c\u6a21\u578b\u4e0b\u8f09\u8207\u7814\u7a76\u8a55\u6e2c\u57fa\u7dda\u4f7f\u7528\uff0c\u76ee\u524d\u53ef\u627e\u5230 BioMatrix-1.7B-Base\u3001BioMatrix-4B-Base\u30011.7B-SFT\u30014B-SFT \u7b49\u7248\u672c\u3002\u82e5\u4f60\u60f3\u6e2c\u8a66\uff0c\u8f03\u5408\u7406\u7684\u7406\u89e3\u65b9\u5f0f\u662f\u5148\u7528\u5df2\u767c\u4f48\u6a21\u578b\u505a\u63a8\u7406\u6216\u4efb\u52d9\u6bd4\u8f03\uff0c\u518d\u6309\u9700\u8981\u7814\u7a76\u5176 tokenizer\uff0c\u4f8b\u5982 MolStructTok \u8207 GCP-VQVAE\uff1b\u5b8c\u6574\u91cd\u8a13\u5c0d\u4e00\u822c\u5718\u968a\u9580\u6abb\u5f88\u9ad8\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u6a21\u578b\u5b9a\u4f4d<\/strong>\uff1a\u591a\u6a21\u614b biological foundation model\uff0c\u4e0d\u662f\u55ae\u4e00\u5206\u5b50\u6a21\u578b\u6216\u55ae\u4e00\u86cb\u767d\u8cea\u6a21\u578b<\/li>\n\n\n\n<li><strong>\u6838\u5fc3\u5dee\u7570<\/strong>\uff1a\u628a sequences\u3001structures\u3001language \u653e\u5165\u540c\u4e00 shared discrete vocabulary\uff0c\u800c\u975e\u9760\u5916\u639b\u5f0f\u6a21\u614b\u6a21\u7d44\u62fc\u63a5<\/li>\n\n\n\n<li><strong>\u76f8\u95dc\u6a21\u578b<\/strong>\uff1aQwen3 1.7B\u3001Qwen3 4B\u3001BioMatrix-1.7B-Base\u3001BioMatrix-4B-Base\u3001BioMatrix-1.7B-SFT\u3001BioMatrix-4B-SFT<\/li>\n\n\n\n<li><strong>\u6578\u64da\u8207\u8a13\u7df4<\/strong>\uff1a\u6db5\u84cb text\u3001PubChem\u3001MolTextNet\u3001UniRef50\u3001RCSB PDB\u3001UniProt\/Swiss-Prot\u3001AFDB \u53ca cross-entity interleaved data<\/li>\n\n\n\n<li><strong>\u8868\u73fe\u6307\u6a19<\/strong>\uff1a\u8ad6\u6587\u7a31 instruction tuning \u5f8c\u6db5\u84cb 80 \u500b tasks\u30016 \u500b\u985e\u5225\uff0c\u7576\u4e2d 77 \u500b tasks \u9054\u5230 state-of-the-art \u6216\u5177\u7af6\u722d\u529b<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u500b\u9805\u76ee\u6700\u53d7\u60e0\u7684\u6703\u662f\u505a drug discovery\u3001protein engineering\u3001\u751f\u7269\u8cc7\u8a0a\u7814\u7a76\uff0c\u6216\u8005\u60f3\u628a\u6587\u5b57\u554f\u7b54\u3001\u5206\u5b50\u8868\u793a\u8207\u7d50\u69cb\u751f\u6210\u653e\u9032\u540c\u4e00\u5de5\u4f5c\u6d41\u7684\u5718\u968a\u3002\u5b83\u7684\u91ce\u5fc3\u5f88\u5927\uff0c\u512a\u52e2\u662f\u7d71\u4e00\u8868\u793a\u8207\u4efb\u52d9\u6cdb\u5316\uff0c\u9650\u5236\u5247\u662f\u90e8\u7f72\u8207\u8a13\u7df4\u9580\u6abb\u9ad8\uff0c\u800c\u4e14\u8ad6\u6587\u8072\u7a31\u7684\u5ee3\u6cdb\u8868\u73fe\u4ecd\u8981\u770b\u4f60\u624b\u4e0a\u7684\u4efb\u52d9\u662f\u5426\u5c6c\u65bc\u90a3 80 \u500b\u6e2c\u8a66\u7bc4\u570d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/QizhiPei\/BioMatrix\" rel=\"noopener noreferrer\">https:\/\/github.com\/QizhiPei\/BioMatrix<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9805\u76ee\u4e3b\u9801\uff1a<\/strong> <a href=\"https:\/\/huggingface.co\/collections\/QizhiPei\/biomatrix\" rel=\"noopener noreferrer\">https:\/\/huggingface.co\/collections\/QizhiPei\/biomatrix<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Paper\uff1a<\/strong> <a href=\"https:\/\/arxiv.org\/pdf\/2606.22138\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/pdf\/2606.22138<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>BioMatrix \u5617\u8a66\u7528\u540c\u4e00\u500b\u6a21\u578b\u540c\u6642\u7406\u89e3\u5206\u5b50\u3001\u86cb\u767d\u8cea\u30013D\u7d50\u69cb\u8207\u6587\u5b57\u3002\u5b83\u7684\u91cd\u9ede\u4e0d\u53ea\u662f\u591a\u6a21\u614b\uff0c\u800c\u662f\u9023\u751f\u6210\u65b9\u5f0f\u90fd\u7d71\u4e00\u8d77\u4f86\u3002<\/p>\n","protected":false},"author":8,"featured_media":9597,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,185,132,131,144,119,76,127,189,200],"tags":[],"class_list":["post-9598","post","type-post","status-publish","format-standard","hentry","category-133","category-qwen","category-3d","category-embedding","category-medical","category-119","category-76","category-127","category-189","category-200"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9598","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=9598"}],"version-history":[{"count":1,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9598\/revisions"}],"predecessor-version":[{"id":9600,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/9598\/revisions\/9600"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/9597"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=9598"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=9598"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=9598"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}